Free medical education that actually works.

Medical education has a tools problem. The resources students actually need to build clinical skills fall into two categories: expensive and bad, or free and worse.

Commercial question banks charge $400+ per year for static multiple choice questions. Heart sound training uses recorded audio clips that sound nothing like what you hear through a stethoscope. EKG courses show you the same 28 stock images. Lecture recordings are 90-minute passive experiences that research consistently shows produce worse retention than active learning. The entire ecosystem is built around memorization, not understanding, and charges a premium for the privilege.

FreeMedEd takes a different approach. Heart sounds are synthesized in real time using the Web Audio API. Change the pathology, the heart rate, or the auscultation maneuver, and hear the difference instantly. There are no recordings. The audio is generated live from physiologically accurate waveform models. EKGs are procedurally generated using Gaussian P-QRS-T synthesis with lead-specific vector projections. Every tracing is unique. Every rhythm is mathematically constructed, not a screenshot from a textbook.

The platform is free. Simulators and course lessons are unlimited. Quiz questions are capped at 10 per day on the free tier, with a $5/month pro tier for unlimited questions, spaced repetition, and exam blueprint scoring. The thesis is simple: if the learning tools are better and the price is zero, the only barrier left is whether students know the platform exists.

The project started as a heart sounds simulator. 28 cardiac pathologies, each modeled as a Web Audio synthesis pipeline: S1 and S2 timing, murmur envelopes, frequency profiles, and gain modeling across four auscultation sites. The phonocardiogram (PCG) waveform renders on a canvas in sync with the audio. Switching between diaphragm and bell modes changes the frequency response in real time. The simulator alone was useful, but it had no structure. You could hear aortic stenosis, but nothing told you why it sounds that way or what to listen for.

That gap produced the Learn mode. 69 lessons across three modules: Foundations (normal heart sounds, physics, physiologic maneuvers), Pathology and Diagnosis (splitting patterns, systolic/diastolic/continuous murmurs), and Clinical Integration (prosthetic valves, integrative reasoning, exam pearls, summary cheat sheets). Every lesson page has learning objectives, sim-trigger cards that launch the relevant pathology in the simulator overlay, and interactive quiz questions. A 6-step interactive Wiggers diagram anchors the physiology.

The EKG course followed the same pattern. 28 rhythms generated procedurally using Gaussian waveform synthesis, rendered as full 12-lead displays on a calibrated pink ECG paper grid. 70 lessons across four modules. The quiz engine uses confusability data to generate smart distractors: if a student misidentifies atrial flutter, the distractor set includes the rhythms most commonly confused with it, not random alternatives.

The concept graph ties everything together. 35 medical concepts linked across courses. Aortic stenosis in the heart sounds course connects to its EKG pattern, its hemodynamics, and eventually its biostatistics (sensitivity/specificity of the murmur as a diagnostic test). The adaptive question engine prioritizes weak concepts, adjusts for recency, and weights questions according to USMLE Step 1 and Medicine Shelf exam blueprints.

The hard problem is synthesis fidelity. A real heart murmur is not a clean waveform. It is shaped by chest wall transmission, stethoscope physics, ambient noise, patient body habitus, and respiratory phase. The Web Audio pipeline models the idealized version: correct frequency content, correct timing relative to the cardiac cycle, correct response to maneuvers. This is pedagogically useful because it isolates the signal, but it means students hear a cleaner version of reality than they will encounter in a clinical setting. The question of how much realism to add before it stops being educational and starts being confusing is an ongoing design tension.

The EKG generator faces a similar tradeoff. Gaussian P-QRS-T synthesis produces clean, readable tracings that highlight the diagnostic features of each rhythm. Real ECGs have baseline wander, muscle artifact, and lead misplacement. Adding noise improves ecological validity but degrades the learning signal for beginners. The current approach is to keep the generated tracings clean and rely on quiz difficulty levels (Order 1 through Order 4) to gradually introduce clinical complexity through the questions, not the tracings.

The adaptive engine is the most architecturally complex piece. It maintains a per-concept mastery estimate, a recency decay function, and an exam-blueprint weighting layer. When a student answers a question, the engine updates mastery for every concept tagged on that question, not just the primary one. Spaced repetition scheduling (SM-2) is in progress. The goal is for the system to know, at any moment, exactly which concept a student should review next to maximize retention per minute of study time.

The biostatistics course is next. 18 units covering sensitivity, specificity, Bayesian reasoning, likelihood ratios, NNT, and study design. The interactive components are partially built: a 2x2 table with real-time calculations, a prevalence slider showing PPV/NPV changes, and a Fagan nomogram for likelihood ratios. The question is whether the adaptive engine can handle cross-domain questions that span auscultation, EKG, and biostatistics simultaneously. The concept graph supports it in theory. Whether the UX supports it without overwhelming students is a different question.

Spaced repetition scheduling needs to ship before the quiz bank becomes large enough for students to lose track of what they have and have not mastered. The SM-2 algorithm is straightforward to implement. The harder problem is integrating it with exam-blueprint weighting so that concepts are prioritized both by memory decay and by exam relevance. A student two weeks from Step 1 should not be reviewing low-yield concepts just because the spaced repetition schedule says so.

The platform currently runs on Supabase for auth and progress tracking. Scaling beyond free tier will require either revenue from pro subscriptions or a migration to a cheaper auth solution. The $5/month price point was chosen to be low enough that cost is never the reason a student does not subscribe, but whether it generates enough revenue to cover infrastructure at scale is an open question.

Pharmacology and pathophysiology courses are on the roadmap. Both require significantly more content authoring than auscultation or EKG, where the simulator can generate infinite practice material. A pharmacology course needs a question bank, and question banks need clinical vignettes, and clinical vignettes need medical accuracy review. The production pipeline for that content does not exist yet.